Thief

The pairing of AMD's Ryzen 9 9900X with NVIDIA's GeForce GTX 1650 yields one of the most lopsided combinations measured in Thief. The CPU is a top-tier component, ranking in the 92nd percentile among all tested processors, while the GPU sits lower, at the 40th percentile. The combination's measured rank of 3595 out of 3722 tested combos places it in the bottom 3.4% — a clear sign that the GPU simply cannot keep pace with the CPU's capabilities. This analysis dissects the measured FPS data, the roles of each component, and what the numbers say about the hopeful pairing.

How This Combo Ranks

The combined benchmark rank for this pairing in Thief is 3595 of 3722, which means only 125 other configurations score lower. That puts it in the bottom 3.4% of the entire list. The data distinctly shows that the CPU handles a High rank alone, with a percentile of 92 among all CPUs and an average benchmark score of 57,498. Meanwhile, the GPU's percentile is 40 with an average score of 7,472 after the game's benchmark suite. The gap between the CPU's near-top-tier standing and the GPU's mid-road state explains why this Title Combo results in such a poor in-game ranking.

A closer look at nearest rivals reinforces the disparity. The Ryzen 9 9900X's closest competitor, the Intel Core i9-14900, has an average score of 58,115, which is a 1.1% higher spot. The GPU's nearest rivals include the Intel Arc A310, which edges it by 1% (avg score 7,550). So the CPU occupies express-range performance territory, while the GPU is at the same level as an integrated graphics part like the Intel UHD Graphics 750 (which is actually 0.4% lower). The traversal of these numbers makes it clear: the CPU is used in the top 8% of processors, but the GPU is in the bottom 70% of graphics cards. When these two combined are respectively bottlenecked, the resulting rank takes a nosedive.

Moreover, the game was originally released in 2014, but the measurement ignores any age influences in the actual output of the combo. The measured data indicates the CPU can feed frames far quicker than the GPU can render. For instance, at 1920x1080 Low, the GPU caps the average to 74 FPS. Even if the CPU could theoretically push more, those numbers stand as the recorded result. The bottom-rank position is thus a direct consequence of the GPU’s performance, not a reflection of the processor's true cap.

GPU Role

The NVIDIA GeForce GTX 1650 is a Turing-generation chip from 2016–2017, with a 4nm process. In terms of its framebuffer, it carries 4 GB of GDDR5 memory over a 128-bit bus, which translates to 128.1 GB/s of memory bandwidth. At the GPU core, the boost clock runs at 1665 MHz (base 1485). The shading units total 896, yielding a FP32 rate of 2.984 TFLOPS. These are modest numbers by today’s standard, and the GPU’s average benchmark score of 7,472 places it in the 40th percentile of all GPUs, which basically means it is below average.

In Thief, the GPU sheds away any hope of a fluid framerate at 4K and delivers only borderline results at lower hardness. At 3840x2160 (4K) with Low settings, the measured average is 30 FPS—already at the threshold of playability. Turn the settings anywhere higher, and the FPS plummet: Medium hits 23 FPS, High and Ultra both drop to 17 FPS. The delta between Low and High shows a 13 FPS penalty, which is a 57% drop in most cases. Such a behavior is typical for a driver that lacks enough render horsepower or memory capacity to handle the resolution load.

The Min/Max data for medium settings highlights the GPU's instability at times. At 1440p Medium, the frame time fluctuates from 33 to 45 FPS, resulting in a span of 12 FPS. At 4K Medium, the range goes from 19 to 26 FPS. That means the GPU is not holding an equal pace, which in many cases indicates that the geometry fill rates are the limiting factor. The GPU's memory bandwidth of 128.1 GB/s also proves to be a choke for pixel data at 4K, especially with hi-def textures, and it's reflected in the drops between Low and other presets. The rendering load is not supported by the 4 GB frame buffer, end of course the GPU always be the determining component in this pairing.

Resolution Scaling

Analyzing across the three measured resolutions (1080p, 1440p, 4K) reveals how FPS declines as the number of pixels grows. Starting with Low settings, the progression is 74 FPS at 1920x1080, 51 FPS at 2560x1440, and 30 FPS at 3840x2160. That's a drop from 74 to 30, or a 59.5% decrease between 1080p and 4K. Meanwhile, the scene's pixel count increases by a factor of about 5.1 in that same span. This relationship is typical of a GPU that is bandwidth-limited and arithmetic-limited at higher resolutions.

At the High and Ultra stockpresets, the FPS is almost the same: 43/41 at 1080p, 29/28 at 1440p, and then 17/17 at 4K. So, once the settings go up, increasing the render load by back up the GPU's design, the curve flattens out between 1080p and 4. While making these trends it suggests that at higher settings the single biggest influence is the explosion in area of pixels, not the setting centering. The CPU's 16 core, 24-thread setup is not needed to spend any effort here; the scenes are not complex enough to tax its multithreaded abilities beyond what at.

In the same sense, the 1440p to 4K transition is particularly harsh: at Low, the FPS goes from 51 to 30 (a 41% shift) and at High it goes from 29 to 17 (a 41% drop). So the proportional loss is almost identical—confirming that the GPU's fill rate and compute capacity are the absolute constraints in that region. The CPU's high computational throughput (Cinebench R23 multicore score is 32,172) doesn't change the fact that the render output count jump overwhelms all. Thus, the scaling pattern is a textbook case of GPU bounded scaling—the GPU is the only component that blocks any performance above these numbers.

Measured FPS Breakdown

For a total picture, here is the complete set of measured frame-rate numbers:

| Resolution | Settings | Avg FPS | Min FPS | Max FPS |

|------------|-----------|---------|---------|---------|

| 1920x1080 | Low | 74 | - | - |

| 1920x1080 | Medium | 56 | 48 | 65 |

| 1920x1080 | High | 43 | - | - |

| 1920x1080 | Ultra | 41 | - | - |

| 2560x1440 | Low | 51 | - | - |

| 2560x1440 | Medium | 39 | 33 | 45 |

| 2560x1440 | High | 29 | - | - |

| 2560x1440 | Ultra | 28 | - | - |

| 3840x2160 | Low | 30 | - | - |

| 3840x2160 | Medium | 23 | 19 | 26 |

| 3840x2160 | High | 17 | - | - |

| 3840x2160 | Ultra | 17 | - | - |

At 1920x1080, the combo's fastest output is 74 FPS on Low, but that falls to 56 if you select Medium. The accrued drop from Low to Ultra is from 74 to 41 (a 45% reduction). With Full HD—which is often the target for e-sports—the GTX 1650 does provide faintly playable playback, but the frame dips hardly get near the 60FPS line unless you drop the settings to Low.

For the 2560x1440 HD, the data show a 51 FPS tune on Low, while Medium secures 39 FPS, High 29, and Ultra 28. This is a clear sign of the increasing effect of the per-pixel shading. The min FPS for 1440p Medium (33) is very close to the average, making that setting somewhat struggle to keep stables. The 29–30 FPS at the upper presets keeps the experience around the 30 period, and there's no headroom.

At 3840x2160, even Low only yields a 30 FPS average. The Medium gives 23 FPS, and the High/Ultra both stuck at 17 FPS. The drop grows exponentially. The above three rows don't include Min/Max for every setting, but the Medium one is 19–26 FPS, showing fluctuation in the 20s. The absolute floor measurement—30 FPS at Low—is barely enough for a fluid reproduction, and any setting of higher stated renders the game 1/6 (?) of normal to the majority user.

CPU Role

The Ryzen 9 9900X is a “Granite Ridge” Zen-5 part, manufactured on a 4 nm process by TSMC. It has 12 physical cores and 24 threads, with a base clock of 4.40 GHz and a boost up to 5.60 GHz. Its 64 MB of L3 cache and 89.6 GB/s dual-channel DDR5 memory bandwidth are respectable at the system level—though none of that matters much in a GPU-bound scenario like Thief. The CPU's average benchmark score of 57,498 places it in the 92% percentile of all CPUs, with a counterpart Intel Core i9-14900 only 1.1% higher score.

In this game, the data clearly shows the CPU is not the limiting resource. At 1920x1080 High settings, the CPU's massive advantage in single-core performance (Geekbench single 3010, Cinebench R23 single 30253) is irrelevant when the GPU is throttling the output to 43 FPS. Conversely, the CPU's memory fabric has 89.6 GB/s, while the GPU's memory bandwidth is 128.1 GB/s—so the memory subsystem aboard the GPU is still more ample than the game's needs, but not the GPU's compute. The CPU is idle 85% of the time, but it does not matter because the GPU can't supply the frames.

The fact that the measured FPS at 1080p Low is 74—and at 4K Ultra is 17—shows that it's the pixel shading, not the CPU logic, that limits